Prof G Markets: Steve Eisman Sees Anthropic's Q3 Slowdown and $800 Billion Commitment Stack as Gaps in an S-1 That Stops at June
Big Short investor says third-quarter data points to the end of "token maxing"; Prof G Markets segment, October 28, 2026
The S-1 Stops Before the Quarter That Matters
Steve Eisman, the investor known for The Big Short and host of The Real Eisman Playbook, told Prof G Markets that the most important feature of Anthropic's S-1 is what it leaves out. The filing's financials run only through the second quarter. Eisman said he has heard Anthropic will have to publish third-quarter numbers before listing, though he is "not 100% sure that's correct." Reports already circulating give third-quarter revenue and compute infrastructure cost, and he argued those figures "tell a tale."
Eisman's framework rests on two statistics: sequential revenue growth and revenue divided by compute infrastructure expense. In the fourth quarter of last year, just before what he called "token maxing" took off, sequential revenue growth was 43% and revenue covered only 0.69x of compute expense. In the first quarter, growth jumped to 135% and the ratio to 1.4x. In the second quarter, the peak, growth reached 147% and the ratio 2.08x. In the third quarter, growth fell to 45% and the ratio slipped to 1.7x. "That tells you that the end of token maxing has had an impact," he said.
Eisman defined token maxing as the period when software engineers were ordered by their employers to use AI around the clock, with little regard for cost. After the first half, he said, customers became far more cost conscious and Chinese open-weight models took share. On his reading, the S-1 is "a snapshot of the company at the height of token maxing" and before Chinese open-weight competition took hold, which makes it backward looking.
Host Ed Elson noted that the charts at the top of the filing leave out the third quarter entirely and that there has been speculation the omission was deliberate. Asked whether he thought so, Eisman answered, "Only very intentional." Elson later said the omission is a theory rather than a fact, and Eisman offered no direct evidence of intent.
A Reported 50% Price Cut and the Question of Moats
Eisman said he has heard, without being able to confirm it, that Anthropic cut prices by roughly 50% in October to win back market share. He treated that as potentially the most important unanswered question before the IPO. "Cutting prices in half could be dangerous," he said, and the questions that follow are how much share was regained and whether revenue rose or fell as a result. He conceded he does not know the answers.
The more consequential point is strategic. If the price cut is real, Eisman said, it reinforces his view that "there are no moats around this business" and that the only way to gain share is price. "Being in a business where you are committed to spend $800 billion, where your entire franchise is just simply based on price, is dangerous." Asked directly whether he would invest, he answered "No."
$800 Billion in Commitments, With Capex Still to Come
Both Eisman and Elson pointed to the future compute commitments disclosed at the end of the filing as the key balance sheet item. Eisman put the total at $800 billion to $900 billion depending on what is included, quipping that at that scale, "what's 100 billion amongst friends?" Elson listed a $110 billion contract with Amazon, a $115 billion contract with Google and a $161 billion contract with Broadcom. None of these flow through the current income statement, though Elson argued all eventually will.
The filing also says Anthropic will significantly raise capital expenditures, which Elson read as a move to build its own data centers. Eisman was more cautious: "I don't know if they're going to build data centers, but they're certainly going to be spending a lot more money. Maybe they'll rent." He cited one report that capex could reach $80 billion to $100 billion next year, and said it is unclear whether it would be expensed or capitalized. Elson noted that the disclosure sits at the bottom of the document "as basically like a footnote," and argued that the $800 billion figure is probably understated and that losses are set to rise sharply. Eisman's own conclusion was narrower: "When you commit to 800 billion dollars, your expenses are going up."
Governance: "You Have No Rights"
Eisman gave the governance structure little time. "You're a shareholder in this company, you have no rights," he said. "It's been disclosed. You want to play? Play." Elson outlined the structure: regular voting and non-voting shares, a Class T share controlled by the Long-Term Benefit Trust that selects directors, and a single Class F share controlled by the seven co-founders that carries 50.1% of the vote.
Eisman did not see this as disqualifying and noted that the same criticism applies to Meta, whose stock has performed well. "This is what goes on in tech world. They don't like their own shareholders." Elson called it institutionalized founder worship.
Valuation and Sentiment
The IPO is expected to price at around $2 trillion, with some debate about a cut to $1.5 trillion. Eisman declined to forecast, saying he has no information beyond press reports. He did suggest the filing probably raised investor confidence, because few people had focused on the second-quarter cutoff. If third-quarter numbers are added before the listing, he said sentiment "should" turn, while acknowledging he can never predict how investors will react.
CoreWeave Debt as the AI Financing Barometer
Beyond Anthropic, Eisman said the item he watches most closely is the price of CoreWeave's debt. At roughly 15%, which he said he cannot prove but believes, borrowing is prohibitive for a company whose model is "we raise debt, we build AI data centers. That's the entire business model." If the cost rises further, he said, "they're really out of business." He allowed that Nvidia could step in and backstop the company, but "barring that," he would be watching the debt markets.
Prof G's Closing Read on the S-1
In a closing segment, Elson separated what he saw as myth from fact. He called the $80 billion first-half net loss a "red herring," attributing it to the fair value increase on convertible notes that converted to equity, a noncash accounting charge. He called first-half revenue growth of 1,300% year over year real but not indicative of the current run rate, given the slowdown from about 150% sequential growth in the second quarter to under 50% in the third. He acknowledged that the narrowing of losses, from more than $2 billion in the first half of last year to under $700 million this year, is a positive, but argued it will not continue.
His conclusion was that the income statement does not tell the whole story. "The business model hasn't been figured out yet," he said, calling Anthropic "a highly risky venture bet" rather than "a proven safe corporate investment, something that its potentially $2 trillion valuation might suggest." He added that "next year, that picture is going to look drastically different."
Autonomous Vehicles: Waymo Leads, Tesla Wins on Cost
The episode also included a segment from Prof G research associate Dan Schulan on autonomous vehicles. He described Waymo as the clear leader, with service in 15 American cities, about 500,000 paid rides a week and roughly seven times Tesla's weekly miles. Tesla's robotaxi business has scaled quickly, with fully commercial service in six cities after removing safety monitors and the recent introduction of its steering-wheel-free Cybercab. Its advantage is cost: Schulan put Cybercab production cost at about $25,000 against more than $100,000 for a Waymo vehicle. About 1.7 million people globally pay for Tesla's consumer FSD software. Amazon's Zoox has about 100 vehicles in two cities, and Uber reaches five cities through partners.
Internationally, Schulan said China leads in low-cost manufacturing while the US leads in commercialization, and the two are unlikely to enter each other's markets. The contest will play out in the Middle East and Europe, which have so far mostly chosen Chinese providers. At home, regulation remains fragmented at the state level, and public opinion is souring: a Gallup poll showed the share of Americans who think autonomous vehicles improve road safety fell 12% from the 52% recorded in 2018.
For investors, Schulan pointed to lidar suppliers, given that 95% of passenger-car lidar sensors are made in China and legislation could restrict them; to fleet maintenance, where Hertz just announced a unit to service autonomous vehicles; and to trucking, where about 44% of marginal costs are driver compensation and benefits. He cited Aurora and Tesla's Semi as exposure, noting the Semi does not yet have full self-driving capability.